Keeping the Record

Ten fields, written down at the time of the trade, decide whether any question in this module can be answered about your own trading. Two of them cannot be recovered afterwards at any price, and a broker statement contains neither.

Prerequisites: Lesson 17, for what p and b are, and lesson 22, which built a career’s worth of risk out of those two and a fraction, and took all three on credit.

Eight trades, and what the ten fields buy

Three weeks of a small record, in two instruments, on an account that stood at $100,000 throughout. Two setups had names before the period started; one trade had no name because it was not in the plan.

When	Setup name	Entry price	Stop price	Position size	Exit price	Why out
5 Jan 09:52	level reclaim	585.20	583.60	600	588.40	target
6 Jan 10:15	level reclaim	587.10	585.80	770	585.80	stop
8 Jan 11:03	level reclaim	512.40	510.20	440	510.05	stop
12 Jan 09:47	range fade	590.00	588.30	580	589.10	manual
13 Jan 13:20	range fade	508.90	506.90	500	513.60	target
15 Jan 10:38	level reclaim	592.50	590.70	550	590.70	stop
16 Jan 14:05	not in the plan	594.20	592.60	1,240	593.10	manual
20 Jan 10:22	range fade	515.30	513.10	455	519.90	target

Nothing below this line was written down. All of it is arithmetic on the table above.

Three of the eight made money, so p is 37.5 per cent. The three winners returned 2.00R, 2.35R and 2.09R, an average of 2.15R. The five losers cost 1.00R, 1.07R, 0.53R, 1.00R and 0.69R, an average of 0.86R — and that number is the first thing the record has told you that no assumption would have. Two of the five were closed by hand before the stop was reached, and one stop filled fifteen cents through its price, at 510.05 against 510.20, which is the slippage of lesson 11 showing up in a real row rather than as a warning.

So the payoff ratio is 2.147 divided by 0.857, which is 2.51, and the expectancy is 0.375 × 2.15 − 0.625 × 0.86, or +0.27R a trade.

Lesson 17 wrote the same thing as E = p × b − (1 − p) and that form gives +0.31 here, which looks like a contradiction and is not. It is the same number in a different unit: lesson 17 measures expectancy in average losses; this measures it in the R you planned, and the record says those are not the same size — the average loss came to 0.857 of a planned R. Multiply 0.314 by 0.857 and you have 0.269 back. The point is not the conversion. It is that only the log knows the two units differ, and lesson 17 had to assume they did not.

Now the same eight trades as the broker sees them. Net +$1,452, which is $181.50 a trade, average winner $2,121, average loser $982, and a payoff ratio of 2.16. That last figure is wrong by about a seventh against the 2.51 the R column gives, and it is wrong for exactly one reason: seven of the eight trades risked between $960 and $1,001, and the eighth risked $1,984. Money mixes the quality of the trades with the size of the bets. R separates them, and separating them is the only reason the stop column exists.

The equity field finishes the job. Divide each trade’s risk by the account it was taken on and seven of the eight sit between 0.96 per cent and 1.00 per cent, which is lesson 20’s rule being followed, visible as a number rather than as a belief. The eighth sits at 1.98 per cent. It is the one trade with no setup name, and it lost. At the size the rule would have given — $1,000 of risk, $1.60 a share, 625 shares — the same trade at the same prices would have lost $687.50 instead of $1,364. The record prices that single decision at $676.50, on a trade that in R terms was one of the smaller losses of the month.

And then the honest part. Eight trades settle nothing at all. Three winners out of eight is 37.5 per cent with a ninety-five per cent interval running from 13.7 per cent to 69.4 per cent, a span of fifty-six points, which by lesson 19 is precisely what a sample this size is worth. Two hundred rows at the same rate would narrow it to thirteen points. That is the whole argument for starting now rather than starting when it matters: the interval closes with rows, the rows only arrive in real time, and there is no version of January you can go back and record in March.

What this does not settle

That the log is complete. Nothing in this scheme obliges you to enter a trade at all, and an omission is not a random draw from the record: whatever made a trade easy to skip is a property of that trade, so the rows most likely to be missing are the ones that would have moved the numbers most. There is one check and it is worth running monthly: the number of closed trades in your log has to equal the number on the statement. That is the single thing the broker can audit for you, and it is worth having for that alone.

That ten fields are enough forever. They are enough for the five questions above. Two later ones need more. Lesson 36 asks what mode the market was in, which is a field you have to record while it is happening because it is contested afterwards. Lesson 72 asks what else was open at the same time, and that one matters here rather than there: every table in this module assumed trades arrive independently, and the field that would let you check whether yours do is not in the ten.

That a pattern in the record is a finding. Cut a record five ways — by day, by hour, by setup, by instrument, by mood — and test each slice at the usual bar, and there is a 22.6 per cent chance that at least one slice looks significant when nothing whatever is wrong. Cut it ten ways and it is 40.1 per cent. Lesson 19’s arithmetic applies to each slice on its own, and every slice is smaller than the record it came from, so the interval on it is wider. A slice is worth acting on when the gap survives that arithmetic and something you can name and change explains it. Without the second half you have found a coincidence you have not met yet.

That the box marked “in the plan” is incorruptible because you tick it first. Ticking it first defeats hindsight, which is a real effect and the reason it is placed there, and it does nothing whatever against the other one: a trader thirty seconds into an impulse is entirely capable of believing the trade is in the plan, and will tick the box honestly and be wrong. The field is only as good as what “the plan” refers to, which is why the setup name has to come from a list fixed in advance and why each name on that list needs its conditions written down. Then the tick stops being a judgement and becomes a lookup: either the trade matches a named setup’s written conditions or it does not. Most people’s setups have never been written to that standard, and until yours are, the fourth row of the ten is the field this lesson is least able to defend.

That the record improves the trading by existing. It does not. It measures, and measurement is retrospective by construction: the state in which the size doubles and the setup has no name is not a state a spreadsheet reaches into. That is lesson 24, and it is the reason this module has one more lesson in it.

The record is the only instrument in this module pointed at you rather than at the market. Two of its ten fields are gone within a week of the trade, and they are the two that say whether you were following your own method or merely present at the time.

Problems

1. Find out what you can no longer recover. Take your last twenty trades and fill in as much of the ten as you can from statements, charts and memory. Mark every cell you cannot fill honestly. Two of the ten cannot come from a statement or a chart at all — the stop as placed and whether the trade was in the plan — so if those columns come back empty, the other eight are worth less than they look.
2. Compute your edge twice. From your own log, work out the win rate, the payoff ratio and the expectancy in R, then work out the same three in currency. If they agree, your sizing has been constant and the two are telling you one thing. If they disagree, the gap is what your sizing decisions have cost or earned you, and it is not part of your edge — it belongs to lesson 20 and it moves the moment you change the rule.
3. Reconcile the count. For the last complete month, count the closed trades in your log and count them on the broker statement. The two numbers should be identical. If they are not, the missing rows are the exercise: write down what they had in common, because whatever that is has been quietly absent from every figure you have computed about yourself.

Sources. Brad M. Barber and Terrance Odean, “Trading Is Hazardous to Your Wealth” (Journal of Finance, 2000), which reconstructs the actual trading of 66,465 households from brokerage records rather than from anything the households reported about themselves, and finds that those who traded most lagged the market by the widest margin — a result that exists only in the record and could not have been arrived at by asking anyone how they were doing. Baruch Fischhoff, “Hindsight ≠ Foresight” (Journal of Experimental Psychology: Human Perception and Performance, 1975), the source for the claim above that knowing the outcome silently revises what you remember expecting, which is why two of the ten fields have to be written before the trade resolves. Halbert White, “A Reality Check for Data Snooping” (Econometrica, 2000), for what happens to a significance level when you test many slices of one record, and for the correction to apply when you want to test a slice properly rather than approximately.

The record tells you afterwards what you did. The next lesson is about the hour in which you do it — the drawdown this module told you to expect, arriving on schedule, with the arithmetic already finished and correct and no help at all.

Educational only. Trading involves substantial risk of loss. Not financial advice. Past performance does not guarantee future results.